CN110462484A

CN110462484A - The structured illumination of illumination geometry structure with optimization

Info

Publication number: CN110462484A
Application number: CN201880019069.5A
Authority: CN
Inventors: 本尼迪克·迪德里希; 罗尔夫·沃特曼; 哈拉尔德·夏德温克尔; 拉尔斯·施托佩
Original assignee: Carl Zeiss Microscopy GmbH
Current assignee: Carl Zeiss Microscopy GmbH
Priority date: 2017-03-31
Filing date: 2018-03-02
Publication date: 2019-11-15
Anticipated expiration: 2038-03-02
Also published as: US20200264419A1; WO2018177680A2; CN110462484B; WO2018177680A3; DE102017106984A1; US11397312B2; EP3602163B1; DE102017106984B4; EP3602163A2; EP4105705A1

Abstract

According to reference measure, the target transfer function (201) of sample object is determined.Then, (250) are optimized according to the optical transfer function (202) of the target transfer function and optical unit, obtains the illumination geometry structure of optimization.

Description

The structured illumination of illumination geometry structure with optimization

Technical field

Generally speaking, various examples of the invention are related to the structured illumination of sample object during image obtains.Of the invention Various examples relate more specifically to determine the preferred illumination geometry structure of structured illumination.

Background technique

In the optical imagery of sample object, the so-called phase contrast image for generating sample object is often valuable 's.In phase contrast image, at least partly picture contrast is to be caused by light by the phase shift of the sample object of imaging 's.Particularly, this makes those not cause the decaying of amplitude or only cause the small decaying of amplitude but have the sample of significant phase shift Object can be imaged with relatively high contrast.In general, such sample object is also referred to as phase object.Biological sample conduct Sample object under microscope, phase change are usually bigger than the amplitude variations of electromagnetic field.

For example, the technology of various phase contrast imagings has: dark-ground illumination, oblique illumination, differential interference contrast's degree (DIC) Or Ze Nike (Zernike) phase contrast.

Above-mentioned technology has various disadvantages or limitation.In general, in the region of so-called detection optical unit, it may be necessary to Additional optical element is provided between sample and detector, in order to phase contrast imaging.This may cause the limitation in structure.

Also know the technology that phase contrast can be obtained by structured illumination.For example, with reference to DE10 2,014 112 242 A1 or US 9,507,138 B2 or L.Tian and L.Waller: " the quantitative differential phase under LED array microscope compares imaging " (Quantitative differential phase contrast imaging in an LED array Microscope), Optics Express 23 (2015), 11394.

However, having certain limitation those described above technology.For example, passing through the illumination geometry structure of fixed setting The phase contrast that can be obtained may be relatively limited.

Summary of the invention

Therefore, it is necessary to improve the technology that sample object is imaged using structured illumination.In particular, needing to be mitigated or eliminated Technology as at least the above some limitations and disadvantage.

The purpose is realized by the feature of independent claims.The characterizing definitions of dependent patent claims embodiment.

A kind of method, comprising: determine object transmitting (transfer) function (otherwise referred to as object transfer of sample object (transmission) function).It here, is that the target transfer function is determined according to reference measure.This method further include: according to The target transfer function, and also according to the optical transfer function of optical unit, optimization is executed, to obtain the illumination of optimization Geometry.This method further include: at least one lighting module is driven, for the illumination geometry structure using the optimization, and And by means of the optical unit, sample object is illuminated.Optionally, this method may include: to drive at least one detector, use In the image for obtaining the sample object by the optical unit, the image is associated with the illumination geometry structure of optimization.

The illumination geometry structure that optimization is obtained by optimizing, even for the illumination mould with multiple freedom degrees Block, such as with multiple adjustable lighting elements, can also effectively search for biggish search space.For example, can completely with Computer based mode carries out the optimization；That is, not needing to consider measurement data simultaneously when optimizing.This allows to spy Do not carry out the optimization rapidly.After this, the illumination geometry structure of the optimization may generate king-sized phase pair Than degree, or may in the associated image of illumination geometry structure from the optimization of the sample object satisfaction it is different fixed The optimisation criteria of justice.Herein, it can save to the described image progress for using the illumination geometry structure of the optimization to obtain Further post-processing step, for example, the latter is combined with other associated images of different illumination geometry structures.Cause This particularly can observe sample object by eyepiece, and phase-contrast is imaged, for example, not doing further number Word post-processing.

A kind of computer program product, including control instruction, the control instruction can be executed by least one processor.Pass through Executing the control instruction makes the processor execute a kind of method.This method comprises: determining that the object of sample object transmits letter Number.Here, the target transfer function is determined according to reference measure.This method further include: letter is transmitted according to the object Number, and also according to the optical transfer function of optical unit, optimization is executed, to obtain the illumination geometry structure of optimization.The party Method further include: at least one lighting module is driven, for the illumination geometry structure using the optimization, and by means of the light Unit is learned, the sample object is illuminated.Optionally, this method may include: to drive at least one detector, for by described Optical unit obtains the image of the sample object, and the image is associated with the illumination geometry structure of the optimization.

A kind of computer program, including control instruction, the control instruction can be executed by least one processor.Pass through execution The control instruction makes the processor execute a method.This method comprises: determining the target transfer function of sample object.This In, the target transfer function is determined according to reference measure.This method further include: according to the target transfer function, and And also according to the optical transfer function of optical unit, optimization is executed, to obtain the illumination geometry structure of optimization.This method is also wrapped It includes: driving at least one lighting module, for the illumination geometry structure using the optimization, and by means of the optics list Member illuminates the sample object.Optionally, this method may include: to drive at least one detector, for passing through the optics Unit obtains the image of the sample object, and the image is associated with the illumination geometry structure of the optimization.

A kind of controller, including at least one processor.At least one described processor is configured to execute following steps: root According to reference measure, the target transfer function of sample object is determined；And according to the target transfer function, and also according to optics The optical transfer function of unit executes optimization, obtains the illumination geometry structure of optimization；And at least one lighting module is driven, is used The sample object is illuminated in the illumination geometry structure using the optimization, and by means of the optical unit.

A kind of method, comprising: obtain the reference picture of sample object.This method further include: according to the reference picture, benefit Employment artificial neural networks (ANN) classify to the sample object.The method also includes: it is determined and is illuminated according to the classification Geometry, and at least one lighting module is driven, for illuminating the sample using the illumination geometry structure of the optimization Object.In addition, the method may include: at least one detector is driven, it is described for obtaining the image of the sample object The image of sample object is associated with the illumination geometry structure of the determination.

Artificial neural network make it possible to by especially efficiently and quickly in a manner of on the basis of the reference picture it is right The sample object carries out suitably trained classification.The result shows that can particularly rapidly be determined suitably by this classification Illumination geometry structure.This can especially promote to apply in real time.Due to when the feature and structure to sample object are classified, by The scalability of the complexity of artificial neural network mapping can also cover king-sized search space-example by this technology Such as, for different types of sample object.

A kind of computer program product, including control instruction, the control instruction can be executed by least one processor.Pass through Executing the control instruction makes the processor execute a method.This method comprises: obtaining the reference picture of sample object.It should Method further include: according to the reference picture, classified using ANN to the sample object.The method also includes: according to The classification determines illumination geometry structure, and drives at least one lighting module, for the illumination geometry using the determination Sample object described in structured illumination.In addition, the method may include: at least one detector is driven, for obtaining the sample The image of this object, the image of the sample object are associated with the illumination geometry structure of the determination.

A kind of computer program, including control instruction, the control instruction can be executed by least one processor.Pass through execution The control instruction makes the processor execute a method.This method comprises: obtaining the reference picture of sample object.This method Further include: according to the reference picture, classified using ANN to the sample object.The method also includes: according to described Classification, determines illumination geometry structure, and drive at least one lighting module, for the illumination geometry structure using the determination Illuminate the sample object.In addition, the method may include: at least one detector is driven, for obtaining the sample pair The image of elephant, the image of the sample object are associated with the illumination geometry structure of the determination.

A kind of controller, including at least one processor.At least one described processor is configured to execute following steps: obtaining Take the reference picture of sample object；According to the reference picture, classified using artificial neural network to the sample object； According to the classification, illumination geometry structure is determined；And at least one lighting module is driven, for the illumination using the determination Geometry illuminates the sample object.

A kind of method, comprising: classify to sample object.Then, it is shone according to the classification access database with determining Bright geometry.This method further include: at least one lighting module is driven, for illuminating using identified illumination geometry structure Sample object.

It for example, can be with the classification of the manually implemented sample object.For example, hand can be carried out to sample object type Dynamic classification, such as biological sample, drinking water, bacterium etc..Then, predefining entry in database accordingly can provide accordingly Link between the sample object of classification and illumination geometry structure appropriate.Particularly, this technology facilitate it is especially accurate and The illumination geometry structure for being suitable for certain general issues is quickly determined, this usually requires to carry out the sample object of same type Imaging.

A kind of computer program product, including control instruction, the control instruction can be executed by least one processor.Pass through Executing the control instruction makes the processor execute a method.This method comprises: classifying to sample object.Then, According to the classification access database to determine illumination geometry structure.In addition, this method further include: at least one is driven to illuminate mould Block for illuminating sample object using identified illumination geometry structure, and optionally drives at least one detector, for obtaining Take the image of sample object associated with the illumination geometry structure.

A kind of computer program, including control instruction, the control instruction can be executed by least one processor.Pass through execution The control instruction makes the processor execute a method.This method comprises: classifying to sample object.Then, according to The classification access database is to determine illumination geometry structure.In addition, this method further include: at least one lighting module is driven, For illuminating sample object using identified illumination geometry structure, and at least one detector is optionally driven, for obtaining The image of sample object associated with the illumination geometry structure.

A kind of controller, including at least one processor.At least one described processor is configured to execute following steps: right Sample object is classified.Then, according to the classification access database to determine illumination geometry structure.In addition, this method is also Include: to drive at least one lighting module, for illuminating sample object using identified illumination geometry structure, and optionally drives At least one detector is moved, for obtaining the image of sample object associated with the illumination geometry structure.

In the case without departing from the scope of protection of the present invention, features described above and features described below not only may be used To use in the corresponding combination of clear stipulaties, and can be used for further combining or being used alone.

Detailed description of the invention

Fig. 1 schematically illustrates the optical system including lighting module, which is configured for various examples In structured illumination is carried out to the sample object fixed by sample rack；

Fig. 2 illustrates lighting module schematically in more detail, wherein the lighting module includes adjustable in various examples Save the matrix of illumination component；

Fig. 3 schematically illustrates the exemplary illumination geometry structure of lighting module according to fig. 2；

Fig. 4 schematically illustrates each side relevant to for obtaining the optimization that is carried out of illumination geometry structure that optimizes Face；

Fig. 5 is the flow chart of an illustrative methods；

Fig. 6 is the flow chart of an illustrative methods；

Fig. 7 is the flow chart of an illustrative methods；

The weighted superposition that Fig. 8 schematically illustrates each component according to multinomial series expansion in various examples is true Fixed exemplary illumination geometry structure；

Fig. 9 is the flow chart of an illustrative methods；

Figure 10 is the flow chart of an illustrative methods.

Specific embodiment

By the description of following exemplary embodiment, the characteristic of foregoing invention, feature and advantage and these characteristics are realized It will become more fully apparent and become apparent from the method for advantage and is understandable and attached drawing is more detailed to the progress of these exemplary embodiments Explanation.

Attached drawing being referred on the basis of preferred embodiment below, the present invention will be described in more detail.In the accompanying drawings, Identical appended drawing reference indicates the same or similar element.These attached drawings are the schematic diagrames of different embodiments of the invention.Institute in figure The element that shows, which is not necessarily to scale, really to be described.On the contrary, different elements shown in figure can reproduce in this way, I.e. their function and general purpose are intelligible to those skilled in the art.Functional unit and member as shown in the figure Connection and coupling between part also can be implemented as being indirectly connected with or coupling.Connection or coupling can be realized in a wired or wireless fashion It closes.Can be by hardware, the combined mode of software or hardware and software realizes functional unit.

Phase contrast imaging technology is as described below.Ernst Abbe formulated illumination geometry structure to light in 1866 Learn the influence of the image creation and contrast method in microscope.According to the law of partially coherent image formation, this document describes utilizations The object, such as cell, fiber etc. with high contrast phase that digital contrast method can not observe bright bright-field microscope The technology of imaging.

Widely used method allows such as Ze Nike phase contrast method or differential interference contrast's degree (DIC) using specific Optical unit to object phase imaging.And technique described herein makes it not need modification camera lens it is achieved that and making It obtains and visible phase object is directly presented in real time when watching by eyepiece.In order to reach this purpose, it needs using especially suitable The illumination geometry structure of conjunction illuminates the sample object.

Various examples described herein are all based on following discovery: in typically application (such as microscope), about sample The prior information of this object is usually no.This is because being often the mesh that people pursue to the imaging of unknown sample object Mark.It means that compared with photoetching technique, for example, (in terms of photoetching not well-known about the prior information of sample object Mask).Nevertheless, various examples described herein make it possible to find suitable illumination geometry structure for sample object.

In some instances, the target transfer function of sample object has been determined according to reference measure.For example, object transmits letter Number can describe influence of the sample object to lighting module incident light.Optical unit is described to imaging with optical transfer function It influences.

Here, there is different technologies to determine target transfer function and/or optical transfer function.It is any to be passed with two-dimensional object The sample object that delivery function t (x, y) is indicated, can resolve into different spatial frequencys with Fourier decomposition.Therefore, each sample This object can be modeled as the superposition of infinite multiple harmonic wave grids.Here, target transfer function t (x, y) can take complexity Value generally uses following form:

T (x, y)=A₀(x, y) e^{I φ (x, y)}=A₀Cos (φ (x, y))+A₀I sin (φ (x, y)),

Wherein, A₀Corresponding amplitude, the complicated phase of φ corresponding objects.Here A₀Specify the damping of incident field amplitude.It compares Under, relative phase delay that phase quantization passes through wave field.It in one example, can be by object according to the model of Abbe Transmission function and optical transfer function combine realization.This technology is also commonly referred to as " sum of light source ", the light that it will give Source is abstracted as the sum of infinite multiple point light sources.The every bit of light source --- itself is relevant, but irrelevant each other --- pass through Plane wave is generated after Fourier inversion at a proper angle.From can be obvious multiplied by complicated target transfer function by incidence wave Find out, the phase offset of the wave generated by off-axis illumination can change Object Spectra.In a frequency domain, the field distribution of light in the picture can To be expressed as the Fourier transformation of optical transfer function and the product of so-called Object Spectra.The Object Spectra also corresponds to pair As transmission function.Single source point coherently illuminates sample object with plane wave, and generates electric field strength in the picture.By to having All light source point summations and subsequent limiting value imitated on light source area consider, by shifting spectrum superposition absolute value square with The point spread function convolution of optical transfer function obtains light intensity.

Another example is related to the target transfer function determined according to the technology of Hopkins；Referring to H.H.Hopkins's " optical imagery diffraction theory " (On the Diffraction Theory of Optical Images), Royal Society's journal A: mathematics, physical engineering science 217 (1953) 408-432.The method for determining target transfer function from Abbe, one of them Expansion light source is deemed to be equivalent to the sum of many mutually incoherent point light sources, according to the partially coherent image formation system of Hopkins Calculating correspond to a simplified approximate variable.Here, after carrying out initial integration on source region, diffraction progression need to only be asked With.It is advantageous in this way, because optical system is separated from calculating.In partially coherent, pair of mapping Linear behavio(u)r is indicated by the superposition of target point pair or the spectrum of target transfer function.Thus, it is possible to determine transmission interaction coefficent square Battle array (TCC), otherwise referred to as partially coherent target transfer function.

TCC corresponds approximately to the transmission function of partially coherent image formation, contains the optical system and illumination geometry knot The attribute of structure.It is limited in the range of the value that TCC is used not equal to 0 by the frequency that optical unit transmits.Therefore, there is high phase The system area of the dry factor or relevant parameters is larger, and wherein TCC ≠ 0, can map higher spatial frequency.TCC is generally included The all information of optical system, and TCC is also commonly considered complex value pupil (complex-valued pupils), such as In Ze Nike phase-contrast, or when such as being triggered by aberration.TCC can promote point of optical transfer function and target transfer function From.In general, TCC is defined as a 4D matrix, wherein each value of 4D matrix should be with individual target transfer function The spatial frequency of Object Spectra is to associated.This value is corresponding with the damping of each frequency pair.By to resulting strong Degree summation, then carries out Fourier inversion, obtains analog image.

TCC can be such that existing optical system stores on computers in the form of 4D matrix, and when change sample object When, multiplication only is carried out with Object Spectra or simulated object transmission function and TCC, rather than as the method for Abbe, pass through FFT Individually propagate each source point.This allows to optimize with particularly efficient calculation.

It in other words, can be by being filtered with the four-dimensional filter function of TCC to the Object Spectra of target transfer function To obtain analog image, wherein TCC can initially be calculated independently of sample object.Therefore, usually further accounting for even can The frequency that can then will not be triggered by sample object.Just because of this, it there may come a time when to need by having resolved into optical system The independent Coherent Part system of quantity is limited to be further processed TCC.This is also generally referred to as the sum of coherent system.In this way, passing through Four-dimensional TCC can be decomposed into its characteristic value and its characteristic function by singular value decomposition.Each characteristic function is corresponding in turn in light source A source point, and generate a dedicated coherence transfer function (kernel function), the weight of the kernel function ultimately generating image When by associated eigenvalue product generate.Generally, due to the energy compensating characteristic of singular value decomposition, characteristic value decaying is very fast.This It results in and has been able to be accurately determined analog image by the superposition realization of several coherent systems.For example, coherence factor S≤ In the case where 0.5, the first kernel is enough, and during the creation of image error less than 10%.For given light Arrangement is learned, even if object is variation, eigenvalue and eigenfunction is also remained unchanged, therefore can equally calculate in advance.Therefore, For example, can be using two dimension transmission interaction coefficent matrix as the approximation or simplification of four-dimensional transmission interaction coefficent matrix.In optical system In the realistic simulation of system, some characteristic functions are only used, wherein four-dimensional transmission interaction coefficent matrix experienced linear decomposition.

Therefore, on the basis of these technologies, target transfer function and optical transfer function can be determined, for example, then One or more analog images are generated according to suitable illumination geometry structure.Then, in some instances, can be passed according to object Delivery function, and then according to optical transfer function, to determine that the illumination geometry structure of optimization optimizes.Here it is possible to according to excellent Change standard, such as based on optical transfer function and target transfer function, iteratively check multiple analog images of sample object.In Here, analog image may be associated from different simulation test illumination geometry structures.

But it is not needed in other examples to determine that the illumination geometry structure of optimization optimizes.For example, according to various Different embodiments can classify to sample object using ANN according to the reference picture obtained before.Then basis point Class determines illumination geometry structure.Relationship between the usual learning sample object of ANN and suitable illumination geometry structure.For this purpose, can With usage data record, usually can also be used as training data record, it provide reference sample object predefined list and Associated preferred reference illumination geometry.

One example implementation of neural network includes convolutional neural networks (CNN).For example, Krizhevsky, Alex, Ilya Sutskever and Geoffrey E.Hinton, " image set of depth convolutional neural networks is classified " (Imagenet Classification with deep convolutional neural networks), neural information processing systems into Exhibition, 2012 or Lawrence, Steve et al., " a kind of recognition of face: convolutional neural networks method " (Face Recognition:A convolutional neural-network approach), IEEE neural network journal 8.1 (1997): 98-113 or Simard, Patrice Y., David Steinkraus and John C.Platt., " convolutional Neural The best practices that network application is analyzed in visible document file " (Best Practices for Convolutional Neural Networks Applied to Visual Document Analysis), ICDAR volumes 3,2003.Here, convolution is in three-dimensional What the so-called convolutional layer between kernel and the three-dimensional sensitizing range (acceptance region) of input matrix (input feature vector figure) determined.At this In, different kernels can be repeatedly applied to the different acceptance regions of input feature vector figure, to provide for being identified The mode of sample object or the translation invariance of feature.

In other examples, manual classification can also be carried out to sample object, wherein can then realize on the database Classification, to be determined as the illumination geometry structure of its function.It may be desirable for such technology, especially with repeat In the routine use of the sample object of type.

Fig. 1 demonstrates an example optical system 100.By example, optical system 100 as shown in Figure 1 be may be implemented Optical microscopy, for example, the optical microscopy with transmitted light geometry.Optical system 100 allows amplification to indicate to fix The small structure of sample object on sample rack 113.For example, a kind of wide visual field microscope may be implemented in optical system 100, at this In kind microscope, sample is illuminated in whole region.For example, by this method, it is examined as sample object thin Born of the same parents' colony can be stared by the eyepiece of the optical unit 112 of optical system 100, to describe it as morbid state；And It does not need computer and entry evaluation or reconstruction is carried out to the data of record.In other examples, optical system 100 can also be realized Laser scanning microscope, in laser scanning microscope, to sample object point by point scanning, and during then, assembling is formed Two dimensional image.

Optical system 100 further includes lighting module 111.Lighting module 111, which is configured to illuminate, to be fixed on sample rack 113 Sample object.For example, can realize the illumination by kohler's illumination.Herein, it uses by condenser and condenser hole Diameter baffle composition.This causes the intensity of light to be for illumination purposes particularly evenly distributed in the plane of sample object.

In the example of fig. 1, lighting module 111 is configured to convenient for structured lighting.It means that utilizing lighting module 111 may be implemented the different illumination geometry structures of the light for illuminating sample object.Herein, in various examples described herein In different technologies can be used to provide different illumination geometry structures.For example, lighting module 111 may include multiple adjustable Illumination component, these adjustable illumination element configurations are locally to modify or issue light.Controller 115 can be with drive lighting module 111 or adjustable element, to realize specific illumination geometry structure.For example, controller 115 can be implemented as microprocessor or micro-control Device processed.Alternatively or additionally as one kind, controller 115 may include such as FPGA or ASIC.

Fig. 2 is shown and 111 related aspect of lighting module.Fig. 2 shows that lighting module 111 includes adjustable in matrix structure The multiplicity of illumination component 121.In other examples, matrix structure can also be replaced with the different geometry arrangement of adjustable element, For example, annular array, semi-circular etc..

In one example, adjustable illumination element 121 can be implemented as light source, for example, as light emitting diode.Then, It is possible, for example, emitting the bright sample object of illumination for the different light emitting diodes with different luminous intensities.It illuminates several What structure can be realized in this way.In further realize, lighting module 111 can be implemented as a space light modulation Device (SLM).SLM can carry out the interference of spatial resolution to optically focused pupil (a condenser pupil), this may have imaging Directly affect-for example, formally being mapped by TCC.Doing so SLM may include multiple adjustable elements 121, for example, micro mirror Or liquid crystal.For example, Digital Micromirror Device (DMD) may be implemented in SLM.Herein, pass through the tiltable mirror of micromechanics mode Son can position on two positions according to the electrostatic field between mirror and carrier material, and electrostatic field can apply from external source. Each adjustable element 121 as pixel may have about 11.8-16 μm of size and the switching frequency of about 5kHz.This is slightly Reflecting mirror deflects into incident beam on absorber or on the direction then applied.In turn, liquid crystal may will affect incidence The phase and/or amplitude of wavefront.The liquid crystal cells that adjustable illumination element 121 can be realized to be arranged between two transparent electrodes. When applying external voltage or extra electric field, the arrangement of these crystal can change.The birefringent characteristic of liquid crystal causes to reflect The spatial variations of rate or the polarization variations of electromagnetic wave.

This SMDs or other SMDs may be provided in the condenser aperture plane of lighting module 111.For example, picture can be used The SLM of elementization replaces condenser aperture diaphragm.By changing the transmission of single pixel, other than any illumination geometry structure, Symmetrical or continuous illumination geometry, for example, annular stops or oblique illumination and possible.In addition, even if with the shape of DMD Formula, SLM can be used for the plane being conjugated with concentrator aperture.Active LED matrix can be equally used for this purpose.In order to ensure just True function, it usually needs there are linear relationships between light source, SLM and camera.The drive of LCD can be calibrated with Gamma correction Dynamic behavior.SLM can be used for transmission mode, and wherein LED or halogen lamp for example represent actual light source.

Fig. 3 is shown and 300 related aspect of example illumination geometry structure.Fig. 3 shows that the various of lighting module 111 can Adjust luminous intensity 301 provided by axis X-X' of the element 121 along Fig. 2.Although Fig. 3 shows the strong light of consecutive variations Degree, but in other examples, lighting module can also provide the illumination component 121 with switching function.

From Fig. 3, it is apparent that adjustable illumination element 121 different in lighting module 111 illuminates geometry shown in Different illumination intensities 301 is provided in structure 300.In various examples described herein, by efficient light sources to strong Degree carries out target masking, or by other implementations of illumination geometry structure, can zoom in or out certain details of sample object Or certain object frequency/items of information.Suitable illumination geometry structure, the photograph can be found in various examples as described herein Frequency caused by bright geometry destroys the phase pair in sample object image due to avoiding or inhibiting destructive interference Degree of ratio.In this way, for example, can obtain particularly preferred image compared with traditional bright field illumination for the image of sample object and imitate Fruit.Other than especially emphasizing phase-contrast, it is also contemplated that other quality standards.

In various examples described herein, the setting of each adjustable illumination element can be optimized.This means that can make Suitable illumination geometry structure is realized with the nonmonotonic or any setting of illumination component.This helps to carry out image result More flexible optimization is especially using the technical aspect such as fixed semicircle illumination geometry structure.For example, from Fig. 3 , it is apparent that illumination geometry structure 300 corresponds to the non-monotonic variation from illumination component 121 to illumination component 121.

Fig. 4 shows and determines 300 related aspect of illumination geometry structure after optimizing by optimization 250.In Fig. 4 institute In the example shown, using computer based algorithm appropriate, it is automatically found the illumination geometry structure 300 of optimization.Here, at this The prior information about existing sample object is not needed in the various technologies of text description.But according to determining pair of reference measure As transmission function 201.Then, 250 are optimized using optical transfer function 202, so as to the illumination geometry knot after being optimized Structure 300.

It, can be according to target transfer function 201, and according to optical transfer function for this purpose, for example, in order to optimize 250 202, iteration determines multiple analog image 211-214 of sample object.Then according to optimisation criteria in analog image 211-214 Certain width image whether meet optimisation criteria and test.For example, different analog image 211-214 may be from different tests Illumination geometry structure is associated.

Then, used optical transfer function 202 can be predetermined, for example, being stored in nonvolatile memory In.For example, optical transfer function 202 may include four-dimensional TCC or two dimension corresponding with the main feature vector of four-dimensional TCC TCC。

Herein, different optical transfer functions 202 can be associated from different illumination geometry structures 300.This meaning One group of optical transfer function 202 can be provided, one group of optical transfer function corresponds to different tests and illuminates geometry knot Structure 300.In other examples, it can also be determined according to the illumination geometry structure for being currently used in corresponding analog image 211-214 Corresponding optical transfer function 202.

From Fig. 4, it is apparent that being used for the parameter of image recording, such as the NA value of object lens and condenser, that is, general next It says, the parameter of optical unit 112 can be a priori known.The mathematical model of optical system can be generated on computers, In, even the model can be calculated only once in the case where changing sample object or target transfer function, then can weigh With, such as TCC.

Fig. 5 illustrates a kind of illustrative methods.Initially, the target transfer function of sample object is determined in 1001.This can It is realized with base according to reference measure.For example, it may be determined that four-dimension TCC or two dimension TCC.

Then, optimization is executed in 1002.This is according to the target transfer function and optical delivery letter determined in 1001 It counts to realize.Optical transfer function can store in a predefined manner in memory.For example, the institute in optimization range The different optical transfer functions for the different test illumination geometry structures considered can store in memory.

Then, the lighting module and detector for obtaining the image of sample object are driven in 1003.Here, using root The illumination geometry structure obtained according to 1002 optimization.It says in principle, the image using detector acquisition sample object is optional 's.Alternatively, for example, sample object only can illuminate with illumination geometry structure obtained, and by suitably illuminating mould Block is realized, observes user in the case where nil, such as pass through eyepiece.

For example, according to that can be realized by the controller 115 of optical system 100 method shown in fig. 5 (see Fig. 1).

Fig. 6 is a kind of flow chart of exemplary method.In order to determine target transfer function, reference measure is carried out in 1011. In different examples described herein, very different technology can be used to determine target transfer function.Correspondingly, required Reference measure may also be different.In order to have the basis of the illumination geometry structure of optimization, it usually needs about sample object Amplitude and phase approximate information item.This means that target transfer function has been determined.Object can be determined by intensity image The amplitude of body.

There are various Phase Build Out algorithms, for example, intensity transmission equation (TIE), passes through iterative process from having along z-axis Different focal point position record storehouse in rebuild phase.This is by record interference pattern and the subsequent other methods assessed Come what is realized.It, can be to rebuild phase based on a variety of records with different light angles by influencing illumination geometry structure Position, for example, by iterative phase algorithm for reconstructing, in another example Gerchberg-Saxton algorithm.

It therefore, can be by drive lighting module 111, to irradiate bright sample object from different initial illumination geometries Reference measure is executed, for example, specific illumination direction, in particular, obtaining the multiple of sample object by driving detector From the different associated initial pictures of initial illumination geometry.Then, according to multiple initial pictures of the sample object, really The phase-contrast weighted reference picture of the fixed sample object.Here, reference picture in some instances may be comprising about right As the qualitative information item of phase.For example, this may be very suitable to be split cell, compare and observe.However, showing at other Example in, can also be determined by object phase with can quantitative assessment item of information phase-contrast weighted reference picture.By This, can determine the refractive index of sample interior.

A kind of technology of phase-contrast weighted reference picture is related to iterative Fourier transform algorithm.Become in iteration Fourier In scaling method, input value is the intensity measurements at some z location, wherein it should find its initial unknown phase, and Further measured value at another z location.Compound light fieldFrom amplitude or the intensity distribution of measurement Square rootWith the initial phase for generally selecting random valueIt is formed.Signal is propagated by Fourier transform Spectrum is generated to frequency space, wherein amplitude is extracted and the amplitude of the spectrum by measuring for the second timeInstead of.This A field travels to previous plane by inverse transformation, and the intensity measured in the plane is again instead of amplitude.The process can have Have in the loop of suitable stop criterion and carry out, reconstructed phase successively converges on its true value.

Fourier's coordinate algorithm also uses similar mode, and wherein z location remains unchanged, and light angle can become Change.The basic thought of the algorithm is sub-light spectrum-shift-being placed on master in frequency space by oblique illumination triggering The corresponding actual position of spectrum.Similar to the variant for iteratively rebuilding phase, phase is by periodically replacing light with measured value It composes and restrains herein.

The step of exemplary Fourier's coordinate analysis, is as follows:

(I) dominant spectral: the high-resolution spectroscopy of amplitude and initial phase from interpolation is generated, for example, factor x2, by force Degree measurementFor example, by being generated based on Fourier transformation from central lighting.

(II) sub-light spectrum is extractedThe frequency range corresponding to i-th of ionization meter is cut out from high-resolution spectroscopy Component.Here, radius r_MOCenter corresponding to the CTF of object lens, and sub-aperture corresponds to

(III) pass through measurement replacement sub-light spectrum: after carrying out inverse Fourier transform to the spectrum of extraction, being surveyed with i-th of intensity The square root of amountInstead of amplitude, phase_LRIt remains unchanged；Before this

(IV) field distribution of update is placed in high-resolution spectroscopy: compound field intensityFrequency is propagated back to by FFT2 Rate space, and after being filtered with CTF, (this is for all 1...n records for the corresponding position being transferred in high-resolution spectroscopy It executes).

(V) convergence of phase: in general, step (II) to (IV) should be carried out about two to three times, until phase convergence is Only.

A kind of relatively simple technology for determining phase contrast weighted reference picture be based on sample object just The combination of beginning image.For example, the relevant art described in 10 2,014 112 242 A1 of DE.

Another kind is to utilize the relationship between phase and intensity image convenient for the method for quantitative assessment phase.This allows to Formulate target transfer function.It is deconvoluted by regularization appropriate to initial data, can quantitatively measure phase.Due to The incoherence in source, limiting frequency are still twice of similar coherent structure.Referring to L.Tian and L.Waller: " LED array Quantitative differential phase under microscope compares imaging ", " optics letter " 23 (2015), 11394.

From example above-mentioned, it is apparent that different technologies can be selected in different realizations to determine that object passes Delivery function.Once it is determined that target transfer function, so that it may continue to optimize.It herein, for example, can be according to reference picture Or target transfer function places the starting point of optimization and/or the boundary condition of optimization.

Fig. 7 is a kind of flow chart of illustrative methods.

Wherein, initial illumination geometry is selected in 1021.This can correspond to rise for the optimal setting then executed Initial point.For example, can be known between previously determined target transfer function and normally suitable initial illumination geometry Relationship is set out, and initial illumination geometry is selected.For example, can be in 1021 using for classifying to target transfer function With the ANN according to the categorizing selection initial illumination geometry.

Then, according to selected optimization algorithm, change current illumination geometry structure in 1022, when front lit geometry knot Structure corresponds to the initial illumination geometry in first time iteration from 1021.

Different optimization algorithms can be used in various examples described herein.For example, particle swarm algorithm can be applied. Particle swarm algorithm does not need to carry out differential (differentiation) to the objective function of description optimisation criteria.

In particle swarm algorithm, the movement modulation corresponding to the single particle of adjustable variable in group's section of lighting module is most Mostly realized by three velocity components:

(I) the proper velocity component (size and Orientation) from previous movement (or initial velocity)；

(II) velocity component of group, i.e., the average value of all particles；And/or

(III) random element corresponding with the decision of biology.

Here, n ties up each particle x-in solution space by source coefficient x_k(1)...x_k(n)-indicate that inverse problem is possible Solution, wherein k is current iterative step.

By the originally determined generated x of target transfer function_k(1)...x_k(n) coefficient random distribution in space or Distribution represents so-called overall good initial value.In each iteration k, all particles of group all experienced repositioning, table Show that repositioning is that occur from the previous position of speed vector sum.

The movement of single particle depends on every other particle, that is, always in an optimal direction, although the latter is different It surely is global minimum value or maximum value.Increment due to each iterative step is changing, different from trellis search method It is that the algorithm can also find optimal between the two nodes；This may be advantageous.

So-called grid-search algorithms are another example optimization algorithms.Grid-search algorithms are that a kind of direct search is calculated Method, the initial parameter distribution of the first step otherwise be based on a set mode, such as isometric net or be it is random or It is according to initial mode intrinsic in solution room.In solution room, with x₁(1)...x₁(i) it is parameter, is with node k The heart, every step iteration is primary, will have a degree of mesh flattening in solution space.Each angle or node k_iIt indicates excellent by what is described The parameter x of the objective function of change standard₁(1)...x₁(i) the new possibility solution indicated.If an objective function node wherein On improved, then the latter will form the center of next iterative step.As long as algorithm is close to minimum value, the range of grid will Reduce.If not finding minimum value in iterative step, increase the range of grid in solution space, with expanded search Region.

The covered search space of optimization compares larger in different examples, optimizes covered search in 1022 Space should the variation appropriate of illuminated geometry covered.This is because usually there are many parameter needs are excellent for lighting module Change, this will affect illumination geometry structure.In one example, for example, optical modulator includes the NxM pixel work in condenser pupil It, can in a digital manner or gray scale switches over for lighting elements.It is generated special due to the different possibility combination of (NxM) ^x Big search space --- wherein x corresponds to the quantity of switchable levels.In general, for example, N and/or M can be > 100, it can also To be > 1000, it might even be possible to be > 5000.

Therefore, the variation for simplifying illumination geometry structure in 1022 is wished in some instances.For this purpose, can be based on multinomial The weighted superpositions of the different components of formula series expansion determines test illumination geometry structure, for example, in some instances, Ze Nike Multinomial.

In other words, therefore the dot structure of the simplified optical modulator of parametrization, or usually available illumination can be passed through Geometry.Come for example, can use zernike polynomial to decomposition circular pupil.Here, zernike polynomial forms radial direction Base.Fenestra can be subdivided into several circle segments.It then, will be each relevant to zernike coefficient using optical transfer function Zernike polynomial or each round section are independently modeled as a test illumination geometry structure in advance, and can store.With different Expression way can predefine the multiplicity of optical transfer function for each component of multinomial series expansion.Then, it utilizes The weighted superposition of the optical transfer function generated in advance is as the superposition for determining analog image.This can substantially reduce calculating expense With.Therefore, can promote to optimize performance faster.

In 1023, analog image can be determined with this technology.Optical transfer function is mutually tied with target transfer function It closes, available analog image.

Then, it tests in 1024 to whether present day analog image meets scheduled optimisation criteria.It can freely select Corresponding mass function is selected, the current state of optimization is quantified.For example, if phase contrast in the foreground, can be simulated again The partially coherent image formation of value object transmission function.When index variation, the contrast of analog image, especially intensity can become Change.Coefficient is modified by optimizer appropriate, minimize it given cost function or improves given optimisation criteria.

For example, optimisation criteria can preferably assess the presence of higher-spatial frequencies in image, for example, this may Lead to higher object detail resolution.Therefore, optimisation criteria may include the picture contrast of analog image.Especially image Contrast can correspond to higher spatial frequency.Optionally or additionally as one kind, optimisation criteria can also include in analog image The image of sample object and the object similarity of reference sample object (fidelity).

Therefore, reference of the reference sample object as during generates optimum angle contrast on this basis.Phase is surveyed Amount, such as phi=0 ... pi, may map to gray value is 0 ... in 255 intensity colors space.Thus, the optimization is grasped Vertical light source, until correspond to the measurement intensity of the result of the target transfer function with TCC convolution or filtering on detector, with from The difference that " phase space " (0 ... pi) is mapped between the object phase of intensity is as small as possible.This is also applied for amplitude, for example, logical Advanced processing amplitude measurement is crossed, low frequency part is cut off, and the image after optimization is compared with the benchmark.

In interferometry field, so-called Michelson contrast are as follows:

With maximum I_maxWith minimum I_minIntensity.For example, since minimum and maximum grey scale pixel value may all be drawn by noise It rises, therefore is dfficult to apply to analog image sometimes.Therefore, it is possible to use CPP contrast, calculates as follows:

So-called " fidelity " or object similarity can be used as further optimisation criteria.This typically refers to ideal image Square of the absolute value of difference between true picture:

F=| | I-I_ideal||²

I_ideal(x)=| t (x) |²or I_ideal(x)=arg (t (x))

In phase object

Transmission function t (x) and Aerial Images I (x) with reconstruction.

If determining that present image is unsatisfactory for optimisation criteria in 1024, to illumination in the new iteration of 1022-1024 Geometry is further changed.Still further, it was discovered that a kind of test illumination geometry structure, with illumination geometry structure one Sample, the practical illumination of the sample suitable for being subsequently used for obtaining image.

Fig. 8 shows the variation that illumination geometry structure is tested in the range of executing optimization.Fig. 8 is illustrated how using ginseng The adjustable light intensity 301 to determine different adjustable elements 121 in lighting module 111 is unfolded in numberization multinomial series.Although Fig. 8 is demonstrated The luminous intensity 301 of one consecutive variations, but in other examples, lighting module can also provide the photograph with switching function Bright element 121.

Fig. 8 shows two components 351 and 353 in exemplary polynomial series expansion.Find optimal illumination geometry knot The freedom degree of structure means a large amount of possible parameters and different combinations.For example, it may be possible to have about 25,000 adjustable illumination member Part 121.Therefore, optimization is limited to advantageous by being typically in the form of for several parameter definitions.By the way that condenser aperture is segmented It is parameterized for zernike polynomial 351,352.Alternatively, circle segments also can be used.Therefore, the quantity of freedom degree sharply subtracts It is few, for example, being reduced to 36 coefficients.Zernike polynomial is that the polar orthogonal expansion for being 1 by radius obtains, therefore is Circle orthogonal polynomial.

It is numbered according to so-called Noll, for example, first calculating the TCC of preceding 37 coefficients, is then folded when optimizing Add.Zernike polynomial is it can also be assumed that be negative value, but light source is not in this way.It would thus be advantageous to the 0 of intensity is replaced with, Such as I₀=0.5, and be I=0..1. by the scope limitation of optimal value.

In order to provide basis for the possibility solution of optimization problem, for example, generating first based on identified target transfer function A illumination geometry structure.From the main diffraction image in pupil overlapping region, Gaussian filter smoothly synthesizes the sharp of plot of light intensity Side, wherein Ze Nike fitting algorithm searches for corresponding zernike coefficient, which represents surface as well as possible.What is obtained is Number indicates the initial parameter of subsequent optimization algorithm.Optimize performance after the completion of, according to in real space link definition function it Between assignment (assignment), can by means of by multinomial series be unfolded as the optimized variable optimized and at least one Multiple discrete adjustable elements 111 of the lighting module of a application, find the illumination geometry structure of optimization.

In conclusion described above is can determine the technology of the illumination geometry structure of optimization by executing optimization.In Here it is possible to predefine the figure obtained followed by the illumination geometry structure found by selecting optimisation criteria appropriate The attribute of picture.Although above with reference to the optimisation criteria (for example, picture contrast or object similarity) in relation to phase contrast Various examples are explained, but also can choose different optimisation criterias in other examples.It is then possible to utilize it is being found, The illumination geometry structure of optimization obtains the image of sample object, and described image has desired characteristic.For example, in order to optimize phase Comparison determines the complex value target transfer function comprising amplitude and phase.

Various algorithms can rebuild phase, or at least with enough formal approximation the latter, without modifying by having The Optical devices of the wide visual field microscope composition of digital optically focused pupil.Fourier's overlapping algorithm illuminates object from multiple directions, and will Different frequency information items is merged into a kind of iterative Fourier transform algorithm (iFTA).Other than increasing space-bandwidth product, this is also Lead to reconstructed phase.

The inverse transfer function of quantitative phase method of comparison provides for the phase of the determining thin opposite elephant for following the first Born approximation One strong selection.Here, in each case, all indicating two half complementary circular patterns on SLM, and will be strong by calculating Image is spent to be combined with each other.

Define random " test " of the parameter of illumination geometry structure --- for example, zernike coefficient-is only in rare cases It brings forth good fruit.Be difficult to analytically calculate gradient since TCC to be used in combination with SVD, generally can not use need to use In the optimization algorithm of the gradient for the cost function for defining optimisation criteria.Therefore, it is possible to use various so-called genetic algorithms.It is so-called The mobile a group particle of particle swarm algorithm, each particle, which ties up solution space by n, indicates the solution with n coefficient.Just as in biology The same in, entire group all follows global maximum.

Fig. 9 is a kind of flow chart of illustrative methods.Fig. 9 shows related to determining that illumination geometry structure is thought based on ANN Aspect.

This technology for determining illumination geometry structure by ANN is to there may come a time when to need uncommon based on once finding Hope the complex process for being avoided determining analog image by means of optical transfer function such as.

In the example depicted in fig. 9, the sample pair being imaged in the reference picture that the algorithm study indicated by ANN had previously obtained As the relationship between preferred illumination geometry structure, 1101.Then, artificial neural network can be according to reference picture to sample Object is classified, and 1102.Also, illumination geometry structure is then determined according to the classification, 1103.Then, it can be used previously Determining illumination geometry structure carrys out the detector of drive lighting module and the image for obtaining sample object, and 1104.From principle On say, the image using detector acquisition sample object is optional.Alternatively, for example, can only utilize found illumination several What structure is realized to illuminate sample object by lighting module appropriate, makes user in without digitized situation It is observed, such as passes through eyepiece.

For example, method shown in Fig. 9 can be realized by the controller 115 of optical system 100.For example, ANN may include CNN.This CNN is particularly suitable for classifying to sample object according to two-dimentional reference picture.

The model table let others have a look at artificial neural networks input and output between correlation, can be on the basis of machine learning It is trained.For example, can be trained according to the reference illumination geometry of reference sample object to artificial neural network.

Herein, the accuracy for giving problem often increases with the size of available training set.For example, if optical system System is only used for metallurgical laboratory, then it may want to train ANN by showing the reference picture of Metal Cutting.In Bioexperiment Room field, it may be necessary to the training ANN on the basis of slice of cell.Can with the illumination geometry structure of dedicated optimization come Teach artificial neural network.For example, during can determining training according to technology relevant to Fig. 4 and Fig. 5 presented hereinbefore Reference illumination geometry.

In some instances, ANN is configurable to provide result in the result space of link definition.In this way, illumination is several What structure can include the assignment to the illumination geometry mechanism of link definition in artificial neural network result space.At this In, certain discretization to available illumination geometry structure can be provided, wherein can distribution accuracy and tightness it Between weighed.

Figure 10 is the flow chart of an exemplary method.Initially, sample object is classified in 1201.As an example, this can With manually implemented.For example, the type of sample object can be set in user.This can be easily achieved, and especially pass through In the case that optical system 100 repeats characterization problems.

Then, on the basis of classification, database is accessed in 1202, for determining illumination geometry structure.Therefore, data Library may include according between the classification in 1201 and the sample object class for being identified as specially suitable illumination geometry structure Assignment.For example, can be according to above with respect to the entry in the creation database of technology described in Fig. 4 and Fig. 5.

Then, drive at least one for illuminating the lighting module of sample object using identified illumination geometry structure, And for obtain with the detector of the image of the associated sample object of illumination geometry structure, 1203.It says in principle, utilizes inspection It is optional for surveying device and obtaining the image of sample object.Alternatively, for example, found illumination geometry knot can be utilized only Structure is realized to illuminate sample object by lighting module appropriate, carries out user in without digitized situation Observation, such as pass through eyepiece.

To sum up, this document describes the technologies that one kind can show the image with phase contrast.This technology can For in different applications.For example, if user directly observes sample object by eyepiece (for example, microscope), it can be with Use this technology.Then, phase contrast is directly visible, for example, not needing further number post-processing.If light The image formation optical unit of system reduces or inhibits the diffraction order for not having positive contribution to target contrast, then further Application field is related to the explanation of phase contrast and object construction.Advantage may include, for example, passing through available video camera Increase available dynamic range, because, for example, the diffraction order for being not equipped with information only results in strength offsets.In addition, these Information is directly to carry out physical measurement with detector to obtain, and is explained rather than just in subsequent last handling process, For example, being deconvoluted by probabilistic algorithm (for example, maximum likelihood method).

Self-evident, the feature and various aspects of the invention of the embodiments of the present invention can be combined with each other.Especially say, These features can be not only used for described combination, can also be used for other combinations without departing from the present invention Or it is used alone.

For example, digital technology described herein can be special by the hardware combined with traditional phase contrast imaging Property and supplemented.For example, technique described herein can be subject to by such as optical element of DIC prism or phase-contrast ring Supplement.Therefore, phase contrast can be improved.For example, this may be because generating in Ze Nike phase contrast sometimes so-called HALO effect, especially because annular diaphragm matches with phase loop and is effectively reduced, i.e. zero level, not by object diffraction Object not will lead to destructive interference and will not generate edge at object edge, this usually makes image evaluation more difficult.

Claims

1. a kind of method, comprising:

According to reference measure, the target transfer function (201) of sample object is determined；

It is executed according to the target transfer function (201), and also according to the optical transfer function of optical unit (112) (202) Optimize (250), obtains the illumination geometry structure (300) of optimization；

Drive at least one lighting module (111) for using the optimization illumination geometry structure (300), and by means of institute It states optical unit (112) and illuminates the sample object.

2. according to the method described in claim 1, further include:

By driving at least one lighting module (111), for illuminating the sample from different initial illumination geometries (300) This object, and by driving at least one detector (114), for obtaining and the different initial illumination geometries (300) multiple initial pictures of the associated sample object, execute the reference measurement；

According to the multiple initial pictures of the sample object, the phase-contrast weighted reference figure of the sample object is determined Picture.

3. according to the method described in claim 2, wherein, according to the combination of the initial pictures of the sample object and/or According to Fourier's lamination imaging analysis and/or according to the combination of iterative Fourier transform algorithm, the institute of the sample object is determined State phase-contrast weighted reference picture.

4. according to the method in claim 2 or 3, further includes:

According to the reference picture, the starting point of the optimization (250) and/or the boundary condition of the optimization (250) are set.

5. method according to any one of claims 1 to 4, wherein

According to optimisation criteria, the optimization (250) iteratively checks that the multiple images of the sample object, described image are based on institute State target transfer function (201) and the optical transfer function (202) simulated, wherein the analog image from it is different Simulation test illumination geometry structure is associated (300).

6. according to the method described in claim 5, wherein,

The optimisation criteria include: the analog image picture contrast and the analog image described in sample object figure Picture and at least one in the object similarity of reference sample object.

7. method according to claim 5 or 6,

According to the weighted superposition of each component of multinomial series expansion, for example, zernike polynomial, determines the test illumination Geometry (300)；

Wherein, optionally, each component for multinomial series expansion is previously determined optical transfer function (202) Multiplicity.

8. method according to any one of claims 1 to 7, further includes:

According in real space link definition, as it is described optimization (250) optimized variable function and described at least one Assignment between the multiplicity of the discrete lighting elements of a lighting module (111) determines the illumination geometry knot of the optimization Structure (300).

9. according to claim 1 to method described in any one of 8, wherein

The optimization (250) includes particle swarm algorithm or grid-search algorithms.

10. according to claim 1 to method described in any one of 9, wherein

The optical transfer function (202) includes four-dimensional transmission interaction coefficent matrix, or transmits interaction coefficent square with the four-dimension The corresponding two dimension transmission interaction coefficent matrix of the principal eigenvector of battle array.

11. a kind of method, comprising:

Obtain the reference picture of sample object；

According to the reference picture, classified using artificial neural network to the sample object；

According to the classification, illumination geometry structure (300) are determined；

At least one lighting module (111) is driven, for illuminating the sample using the illumination geometry structure (300) of the determination Object.

12. according to the method for claim 11, wherein the neural network is convolutional neural networks.

13. method according to claim 11 or 12, wherein the determination illumination geometry structure (300) includes pair Illumination geometry structure (300) assignment of link definition in the result space of the neural network.

14. method described in any one of 1 to 13 according to claim 1,

According to the reference illumination geometry (300) of reference sample object, the training neural network.

15. a kind of method, comprising:

Classify to sample object；

According to the classification access database to determine illumination geometry structure (300)；

At least one lighting module (111) is driven, for illuminating the sample pair using identified illumination geometry structure (300) As.

16. a kind of controller, including at least one processor, for executing following steps:

According to the target transfer function (201), and also according to the optical transfer function of optical unit (112) (202)；It executes Optimize (250) to obtain the illumination geometry structure (300) of optimization；

At least one lighting module (111) is driven, for the illumination geometry structure (300) using the optimization and by means of institute It states optical unit (112) and illuminates the sample object.

17. a kind of controller, including at least one processor, for executing following steps:

Obtain the reference picture of sample object；

18. a kind of controller, including at least one processor, for executing following steps:

Classify to sample object；